项目名称: 删失数据中位数回归模型的统计分析
项目编号: No.11201235
项目类型: 青年科学基金项目
立项/批准年度: 2013
项目学科: 数理科学和化学
项目作者: 周秀轻
作者单位: 南京师范大学
项目金额: 22万元
中文摘要: 对删失数据回归模型而言,由于因变量的观测出现偏差,使用中位数作为因变量的"中心"的估计比使用均值要简单有效,因此,删失数据中位数回归模型成为均值回归模型的有效替代和补充。本项目将考虑因变量为删失数据时的中位数回归模型,基于最小一乘(LAD)的思想,从以下几方面展开研究:首先考虑因变量含不同类型的区间删失数据时的线性中位数回归模型的LAD估计问题,提出估计方法并证明其渐近性质;基于该LAD估计,分别使用经验似然、随机加权和bootstrap法讨论模型中参数的区间估计和假设检验等统计推断问题;分别对因变量为右删失、双侧删失或区间删失数据时的EV线性中位数回归模型,考虑参数的LAD型估计,以及基于该估计量的区间估计、假设检验等统计推断;分别对因变量为双侧删失或区间删失数据的线性中位数回归模型,使用罚函数的方法考虑变量选择问题。本研究将进一步完善删失数据回归模型的理论,极大的推进该模型的实际应用。
中文关键词: 删失数据;分位数回归模型;最小一乘估计;渐近性质;稳健性
英文摘要: Usually, the median, compared with the mean, is more simple and efficient measure for the center of variables which are censored. Therefore, the censored median regression model provides a valuable complement to censored regression model with zero-mean errors. In the study of this project, we will focus ours most attention on the median regression models with censored responses. Based on the method of Least absolute deviations (LAD for short), the problems will be studied as follows. Firstly, we will propose LAD estimate for linear median regression models with interval censored responses and prove the asymptotic properties of the estimators. Secondly, using the LAD estimators, statistical inference based on empirical likelihood, random weighting, and bootstrap, respectivelly, will be studied. Then, we will study the EV median regression models with right censoring, doubly censoring or interval censoring respectivelly, and propose LAD type methods. Finally, penalty methods will be used to make model selection for linear median regression models with doubly or interval censored responses. This research will further improve the theories of censored regression model, and promote practical application of the model.
英文关键词: Censored data;Quantile regression model;LAD estimation;Asymptotic property;Robustness